I forget which early GPU-labelled card I got but when I got it, I aligned myself with nvidia (as a graphics provider) because it was clear that the pipelines could be re-used for other purposes (in my case, I wanted to run molecular dynamics simulations). The card replaced the previous one (ATI FireGL which had excellent OpenGL support for Linux early on) and I've stuck exclusively with nVidia since. They are the only company I've seen that saw the early vision and carried it through with focus through today. Nowadays I run a large fleet of nvidia GPUs for frontier research at work and also have the last few generations of cards in workstations at home.